move some test from test_matrixCompare.cpp to test_BaseMatrix.cpp and test_Matrix.cpp

paddle/math/BaseMatrix.cu

@@ -1578,11 +1578,6 @@ void BaseMatrixT<real>::minRows(BaseMatrixT& b) {
   applyRow(aggregate::min(), b);
 }
 
-template<>
-void BaseMatrixT<real>::sumCols(BaseMatrixT& b) {
-  applyCol(aggregate::sum(), b);
-}
-
 template<>
 void BaseMatrixT<real>::maxCols(BaseMatrixT& b) {
   applyCol(aggregate::max(), b);

paddle/math/BaseMatrix.h

@@ -1007,8 +1007,6 @@ public:
   /// calculate the minimum value of each row of the matrix b.
   void minRows(BaseMatrixT& b);
 
-  /// calculate the sum of each column of the matrix b.
-  void sumCols(BaseMatrixT& b);
   /// calculate the maximum value of each column of the matrix b.
   void maxCols(BaseMatrixT& b);
   /// calculate the minimum value of each column of the matrix b.

paddle/math/tests/TensorCheck.h

@@ -110,4 +110,10 @@ void TensorCheck(AssertEq compare, real args1, real args2) {
                                          << ", args2 = " << args2;
 }
 
+template <typename AssertEq>
+void TensorCheck(AssertEq compare, size_t args1, size_t args2) {
+  EXPECT_EQ(args1, args2) << "[Test error] args1 = " << args1
+                          << ", args2 = " << args2;
+}
+
 }  // namespace autotest

paddle/math/tests/TestUtils.h

@@ -65,15 +65,24 @@ public:
 // construct a argument
 template <typename T>
 T construct(int height, int width);
+
 template <>
 float construct(int height, int width) {
   return 0.0;
 }
+
+template <>
+size_t construct(int height, int width) {
+  size_t offset = std::rand() % (height < width ? height : width);
+  return offset;
+}
+
 template <>
 CpuMatrix construct(int height, int width) {
   CpuMatrix a(height, width);
   return a;
 }
+
 template <>
 GpuMatrix construct(int height, int width) {
   GpuMatrix a(height, width);
@@ -83,14 +92,22 @@ GpuMatrix construct(int height, int width) {
 // init a argument
 template <typename T>
 void init(T& v);
+
 template <>
 void init(float& v) {
   v = 0.5;
 }
+
+template <>
+void init(size_t& v) {
+  return;
+}
+
 template <>
 void init(CpuMatrix& v) {
   v.randomizeUniform();
 }
+
 template <>
 void init(GpuMatrix& v) {
   v.randomizeUniform();
@@ -111,10 +128,17 @@ template <std::size_t I = 0, typename... Args>
 // copy a argument, copy src to dest
 template <typename T1, typename T2>
 void copy(T1& dest, T2& src);
+
 template <>
 void copy(float& dest, float& src) {
   dest = src;
 }
+
+template <>
+void copy(size_t& dest, size_t& src) {
+  dest = src;
+}
+
 template <>
 void copy(GpuMatrix& dest, CpuMatrix& src) {
   dest.copyFrom(src);
@@ -165,8 +189,8 @@ R call(C& obj, R (FC::*f)(FArgs...), Args&&... args) {
   return (obj.*f)(args...);
 }
 
-template <bool ApplyRow,
-          bool ApplyCol,
+template <bool AsRowVector,
+          bool AsColVector,
           std::size_t... I,
           typename C,
           typename R,
@@ -177,8 +201,8 @@ void BaseMatrixCompare(R (C::*f)(Args...),
                        bool checkArgs = false) {
   for (auto height : {1, 11, 73, 128, 200, 330}) {
     for (auto width : {1, 3, 32, 100, 512, 1000}) {
-      CpuMatrix obj1(ApplyCol ? 1 : height, ApplyRow ? 1 : width);
-      GpuMatrix obj2(ApplyCol ? 1 : height, ApplyRow ? 1 : width);
+      CpuMatrix obj1(AsRowVector ? 1 : height, AsColVector ? 1 : width);
+      GpuMatrix obj2(AsRowVector ? 1 : height, AsColVector ? 1 : width);
       init(obj1);
       copy(obj2, obj1);
 
@@ -227,7 +251,7 @@ void BaseMatrixCompare(R (C::*f)(Args...), bool checkArgs = false) {
 }
 
 template <std::size_t... I, typename C, typename R, typename... Args>
-void BaseMatrixApplyRow(R (C::*f)(Args...)) {
+void BaseMatrixAsColVector(R (C::*f)(Args...)) {
   static_assert(sizeof...(I) == sizeof...(Args),
                 "size of parameter packs are not equal");
 
@@ -237,11 +261,11 @@ void BaseMatrixApplyRow(R (C::*f)(Args...)) {
   autotest::AssertEqual compare(1e-8);
 #endif
 
-  autotest::BaseMatrixCompare<true, false, I...>(f, compare);
+  autotest::BaseMatrixCompare<false, true, I...>(f, compare);
 }
 
 template <std::size_t... I, typename C, typename R, typename... Args>
-void BaseMatrixApplyCol(R (C::*f)(Args...)) {
+void BaseMatrixAsRowVector(R (C::*f)(Args...)) {
   static_assert(sizeof...(I) == sizeof...(Args),
                 "size of parameter packs are not equal");
 
@@ -250,5 +274,5 @@ void BaseMatrixApplyCol(R (C::*f)(Args...)) {
 #else
   autotest::AssertEqual compare(1e-8);
 #endif
-  autotest::BaseMatrixCompare<false, true, I...>(f, compare);
+  autotest::BaseMatrixCompare<true, false, I...>(f, compare);
 }

paddle/math/tests/test_BaseMatrix.cpp

@@ -24,7 +24,6 @@ limitations under the License. */
 #include "TestUtils.h"
 
 using namespace paddle;  // NOLINT
-using namespace std;     // NOLINT
 
 /**
  * Test member functions which prototype is
@@ -32,7 +31,7 @@ using namespace std;     // NOLINT
  */
 TEST(BaseMatrix, void) {
   typedef void (BaseMatrix::*FunctionProto)();
-  #define BASEMATRIXCOMPARE(function) \
+#define BASEMATRIXCOMPARE(function) \
   BaseMatrixCompare(static_cast<FunctionProto>(&BaseMatrix::function));
 
   BASEMATRIXCOMPARE(neg);
@@ -46,7 +45,7 @@ TEST(BaseMatrix, void) {
   BASEMATRIXCOMPARE(zero);
   BASEMATRIXCOMPARE(one);
 
-  #undef BASEMATRIXCOMPARE
+#undef BASEMATRIXCOMPARE
 }
 
 /**
@@ -55,7 +54,7 @@ TEST(BaseMatrix, void) {
  */
 TEST(BaseMatrix, real) {
   typedef void (BaseMatrix::*FunctionProto)(real);
-  #define BASEMATRIXCOMPARE(function) \
+#define BASEMATRIXCOMPARE(function) \
   BaseMatrixCompare<0>(static_cast<FunctionProto>(&BaseMatrix::function));
 
   BASEMATRIXCOMPARE(pow);
@@ -67,7 +66,7 @@ TEST(BaseMatrix, real) {
   BASEMATRIXCOMPARE(biggerThanScalar);
   BASEMATRIXCOMPARE(downClip);
 
-  #undef BASEMATRIXCOMPARE
+#undef BASEMATRIXCOMPARE
 }
 
 /**
@@ -76,13 +75,13 @@ TEST(BaseMatrix, real) {
  */
 TEST(BaseMatrix, real_real) {
   typedef void (BaseMatrix::*FunctionProto)(real, real);
-  #define BASEMATRIXCOMPARE(function) \
+#define BASEMATRIXCOMPARE(function) \
   BaseMatrixCompare<0, 1>(static_cast<FunctionProto>(&BaseMatrix::function));
 
   BASEMATRIXCOMPARE(add);
   BASEMATRIXCOMPARE(clip);
 
-  #undef BASEMATRIXCOMPARE
+#undef BASEMATRIXCOMPARE
 }
 
 /**
@@ -91,7 +90,7 @@ TEST(BaseMatrix, real_real) {
  */
 TEST(BaseMatrix, BaseMatrix) {
   typedef void (BaseMatrix::*FunctionProto)(BaseMatrix&);
-  #define BASEMATRIXCOMPARE(function) \
+#define BASEMATRIXCOMPARE(function) \
   BaseMatrixCompare<0>(static_cast<FunctionProto>(&BaseMatrix::function));
 
   BASEMATRIXCOMPARE(assign);
@@ -129,7 +128,7 @@ TEST(BaseMatrix, BaseMatrix) {
   BASEMATRIXCOMPARE(addP2P);
   BASEMATRIXCOMPARE(invSqrt);
 
-  #undef BASEMATRIXCOMPARE
+#undef BASEMATRIXCOMPARE
 }
 
 /**
@@ -138,7 +137,7 @@ TEST(BaseMatrix, BaseMatrix) {
  */
 TEST(BaseMatrix, BaseMatrix_real) {
   typedef void (BaseMatrix::*FunctionProto)(BaseMatrix&, real);
-  #define BASEMATRIXCOMPARE(function) \
+#define BASEMATRIXCOMPARE(function) \
   BaseMatrixCompare<0, 1>(static_cast<FunctionProto>(&BaseMatrix::function));
 
   BASEMATRIXCOMPARE(addBias);
@@ -154,7 +153,7 @@ TEST(BaseMatrix, BaseMatrix_real) {
 
   BASEMATRIXCOMPARE(isEqualTo);
 
-  #undef BASEMATRIXCOMPARE
+#undef BASEMATRIXCOMPARE
 }
 
 /**
@@ -163,7 +162,7 @@ TEST(BaseMatrix, BaseMatrix_real) {
  */
 TEST(BaseMatrix, BaseMatrix_BaseMatrix) {
   typedef void (BaseMatrix::*FunctionProto)(BaseMatrix&, BaseMatrix&);
-  #define BASEMATRIXCOMPARE(function) \
+#define BASEMATRIXCOMPARE(function) \
   BaseMatrixCompare<0, 1>(static_cast<FunctionProto>(&BaseMatrix::function));
 
   BASEMATRIXCOMPARE(softCrossEntropy);
@@ -181,69 +180,25 @@ TEST(BaseMatrix, BaseMatrix_BaseMatrix) {
   BASEMATRIXCOMPARE(dotMulSquare);
   BASEMATRIXCOMPARE(dotSquareSquare);
 
-  #undef BASEMATRIXCOMPARE
+#undef BASEMATRIXCOMPARE
 }
 
-/**
- * Test aggregate member functions which prototype is
- * void (BaseMatrix::*)(BaseMatrix&).
- */
-TEST(Aggregate, BaseMatrix) {
-  typedef void (BaseMatrix::*FunctionProto)(BaseMatrix&);
-  #define BASEMATRIXAPPLYROW(function) \
-    BaseMatrixApplyRow<0>(static_cast<FunctionProto>(&BaseMatrix::function));
-
- #define BASEMATRIXAPPLYCOL(function) \
-    BaseMatrixApplyCol<0>(static_cast<FunctionProto>(&BaseMatrix::function));
-
-  BASEMATRIXAPPLYROW(maxRows);
-  BASEMATRIXAPPLYROW(minRows);
-
-  BASEMATRIXAPPLYCOL(sumCols);
-  BASEMATRIXAPPLYCOL(maxCols);
-  BASEMATRIXAPPLYCOL(minCols);
-
-  #undef BASEMATRIXAPPLYROW
-  #undef BASEMATRIXAPPLYCOL
+// member function without overloaded
+TEST(BaseMatrix, Other) {
+  BaseMatrixCompare<0, 1, 2>(&BaseMatrix::rowScale);
+  BaseMatrixCompare<0, 1, 2>(&BaseMatrix::rowDotMul);
+  BaseMatrixCompare<0, 1, 2, 3>(&BaseMatrix::binaryClassificationError);
 }
 
-/**
- * Test aggregate member functions which prototype is
- * void (BaseMatrix::*)(BaseMatrix&, BaseMatrix&).
- */
-TEST(Aggregate, BaseMatrix_BaseMatrix) {
-  typedef void (BaseMatrix::*FunctionProto)(BaseMatrix&, BaseMatrix&);
-  #define BASEMATRIXAPPLYROW(function) \
-    BaseMatrixApplyRow<0, 1>(static_cast<FunctionProto>(&BaseMatrix::function));
-
-  #define BASEMATRIXAPPLYCOL(function) \
-    BaseMatrixApplyCol<0, 1>(static_cast<FunctionProto>(&BaseMatrix::function));
-
-  BASEMATRIXAPPLYCOL(addDotMulVMM);
-
-  #undef BASEMATRIXAPPLYROW
-  #undef BASEMATRIXAPPLYCOL
-}
-
-/**
- * Test aggregate member functions which prototype is
- * void (BaseMatrix::*)(BaseMatrix&, real, real).
- */
-TEST(Aggregate, BaseMatrix_real_real) {
-  typedef void (BaseMatrix::*FunctionProto)(BaseMatrix&, real, real);
-  #define BASEMATRIXAPPLYROW(function) \
-    BaseMatrixApplyRow<0, 1, 2>(\
-      static_cast<FunctionProto>(&BaseMatrix::function));
-
-  #define BASEMATRIXAPPLYCOL(function) \
-    BaseMatrixApplyCol<0, 1, 2>(\
-      static_cast<FunctionProto>(&BaseMatrix::function));
-
-  BASEMATRIXAPPLYROW(sumRows);
-  BASEMATRIXAPPLYCOL(sumCols);
+TEST(BaseMatrix, Aggregate) {
+  BaseMatrixAsColVector<0>(&BaseMatrix::maxRows);
+  BaseMatrixAsColVector<0>(&BaseMatrix::minRows);
+  BaseMatrixAsColVector<0, 1, 2>(&BaseMatrix::sumRows);
 
-  #undef BASEMATRIXAPPLYROW
-  #undef BASEMATRIXAPPLYCOL
+  BaseMatrixAsRowVector<0>(&BaseMatrix::maxCols);
+  BaseMatrixAsRowVector<0>(&BaseMatrix::minCols);
+  BaseMatrixAsRowVector<0, 1>(&BaseMatrix::addDotMulVMM);
+  BaseMatrixAsRowVector<0, 1, 2>(&BaseMatrix::sumCols);
 }
 
 int main(int argc, char** argv) {

paddle/math/tests/test_Matrix.cpp

@@ -19,25 +19,20 @@ limitations under the License. */
  */
 
 #include <gtest/gtest.h>
-#include "paddle/utils/Util.h"
-#include "paddle/math/BaseMatrix.h"
 #include "TestUtils.h"
 
 using namespace paddle;  // NOLINT
-using namespace std;     // NOLINT
 
-/**
- * Test member functions which prototype is
- * void (Matrix::*)(Matrix&).
- */
-TEST(BaseMatrix, real) {
-  typedef void (Matrix::*FunctionProto)(Matrix&);
-#define MATRIXCOMPARE(function) \
-  BaseMatrixCompare<0>(static_cast<FunctionProto>(&Matrix::function), true);
+TEST(Matrix, Matrix) {
+  BaseMatrixCompare<0>(&Matrix::softmax, true);
+  BaseMatrixCompare<0, 1>(&Matrix::sumOfSquaresBp);
+}
 
-  MATRIXCOMPARE(softmax);
+TEST(Matrix, Aggregate) {
+  BaseMatrixAsRowVector<0, 1>(
+      static_cast<void (Matrix::*)(Matrix&, real)>(&Matrix::collectBias));
 
-#undef MATRIXCOMPARE
+  BaseMatrixAsColVector<0, 1>(&Matrix::sumOfSquares);
 }
 
 int main(int argc, char** argv) {

paddle/math/tests/test_matrixCompare.cpp

@@ -448,60 +448,6 @@ void testMatrixZeroAtOffset(int height, int width) {
   MatrixCheckEqual(*cpuA, *cpuTest);
 }
 
-void testMatrixSumOfSquaresBp(int height, int width) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuC = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
-  MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
-  MatrixPtr gpuC = std::make_shared<GpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  cpuB->randomizeUniform();
-  cpuC->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  gpuB->copyFrom(*cpuB);
-  gpuC->copyFrom(*cpuC);
-
-  cpuA->sumOfSquaresBp(*cpuB, *cpuC);
-  gpuA->sumOfSquaresBp(*gpuB, *gpuC);
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckErr(*cpuA, *outputCheck);
-}
-
-void testMatrixBinaryRowScale(int height, int width) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, 1);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
-  MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, 1);
-
-  MatrixPtr cpuA1 = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuB1 = std::make_shared<CpuMatrix>(height, 1);
-  MatrixPtr gpuA1 = std::make_shared<GpuMatrix>(height, width);
-  MatrixPtr gpuB1 = std::make_shared<GpuMatrix>(height, 1);
-
-  cpuA->randomizeUniform();
-  cpuB->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  gpuB->copyFrom(*cpuB);
-  cpuA1->copyFrom(*cpuA);
-  cpuB1->copyFrom(*cpuB);
-  gpuA1->copyFrom(*cpuA);
-  gpuB1->copyFrom(*cpuB);
-
-  cpuA->addColVector(*cpuB);
-  gpuA->addColVector(*gpuB);
-  cpuA1->addColumnVector(*cpuB1);
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckEqual(*cpuA, *outputCheck);
-
-  MatrixCheckEqual(*cpuA, *cpuA1);
-}
-
 void testMatrixAddBias(int height, int width, real scale) {
   MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
   MatrixPtr cpuB = std::make_shared<CpuMatrix>(1, width);
@@ -521,76 +467,6 @@ void testMatrixAddBias(int height, int width, real scale) {
   MatrixCheckErr(*cpuA, *outputCheck);
 }
 
-void testMatrixTernaryRowScale(int height, int width) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuC = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
-  MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
-  MatrixPtr gpuC = std::make_shared<GpuMatrix>(height, width);
-
-  MatrixPtr cpuA1 = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuB1 = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuC1 = std::make_shared<CpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  cpuB->randomizeUniform();
-  cpuC->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  gpuB->copyFrom(*cpuB);
-  gpuC->copyFrom(*cpuC);
-  cpuA1->copyFrom(*cpuA);
-  cpuB1->copyFrom(*cpuB);
-  cpuC1->copyFrom(*cpuC);
-
-  int columnOffset = rand() % width;  // NOLINT
-
-  cpuA->rowScale(columnOffset, *cpuB, *cpuC);
-  gpuA->rowScale(columnOffset, *gpuB, *gpuC);
-  cpuA1->rowScale2(columnOffset, *cpuB1, *cpuC1);
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckEqual(*cpuA, *outputCheck);
-
-  MatrixCheckEqual(*cpuA, *cpuA1);
-}
-
-void testMatrixTernaryRowDotMul(int height, int width) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuC = std::make_shared<CpuMatrix>(height, width);
-
-  MatrixPtr cpuA1 = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuB1 = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuC1 = std::make_shared<CpuMatrix>(height, width);
-
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
-  MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
-  MatrixPtr gpuC = std::make_shared<GpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  cpuB->randomizeUniform();
-  cpuC->randomizeUniform();
-  cpuA1->copyFrom(*cpuA);
-  cpuB1->copyFrom(*cpuB);
-  cpuC1->copyFrom(*cpuC);
-  gpuA->copyFrom(*cpuA);
-  gpuB->copyFrom(*cpuB);
-  gpuC->copyFrom(*cpuC);
-
-  int columnOffset = rand() % width;  // NOLINT
-
-  cpuA->rowDotMul(columnOffset, *cpuB, *cpuC);
-  gpuA->rowDotMul(columnOffset, *gpuB, *gpuC);
-  cpuA1->rowDotMul2(columnOffset, *cpuB1, *cpuC1);
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckErr(*cpuA, *cpuA1);
-  MatrixCheckErr(*cpuA, *outputCheck);
-}
-
 void testMatrixAddDotMulMMV(int height, int width) {
   MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
   MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
@@ -670,18 +546,11 @@ TEST(Matrix, unary) {
     for (auto width : {1, 3, 32, 100, 512, 1000, 3210}) {
       VLOG(3) << " height=" << height << " width=" << width;
 
-      // applyTernary
-      testMatrixSumOfSquaresBp(height, width);
-
       // asRowVector
       testMatrixAddBias(height, width, 1.0);
       testMatrixAddBias(height, width, 3.5);
       testMatrixAddDotMulMMV(height, width);
 
-      // asColVector
-      testMatrixTernaryRowScale(height, width);
-      testMatrixBinaryRowScale(height, width);
-
       // sum
       testMatrixGetSum(height, width);
 
@@ -782,119 +651,6 @@ TEST(Matrix, softmax) {
   }
 }
 
-void testMatrixCollectBias(int height, int width) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(1, width);
-  MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(1, width);
-  MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  cpuB->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  gpuB->copyFrom(*cpuB);
-
-  real scale = 1.0f / (rand() % 10);  // NOLINT
-
-  cpuA->collectBias(*cpuB, scale);
-  gpuA->collectBias(*gpuB, scale);
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(1, width);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckErr(*cpuA, *outputCheck);
-}
-
-void testMatrixSumOfSquares(int height, int width, int endCol = 0) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, 1);
-  MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuC = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, 1);
-  MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
-  MatrixPtr gpuC = std::make_shared<GpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  cpuB->randomizeUniform();
-  cpuC->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  gpuB->copyFrom(*cpuB);
-  gpuC->copyFrom(*cpuC);
-
-  if (!endCol) {
-    cpuA->sumOfSquares(*cpuB, *cpuC);
-    gpuA->sumOfSquares(*gpuB, *gpuC);
-  } else {
-    MatrixPtr subCpuB = cpuB->subColMatrix(0, endCol);
-    MatrixPtr subCpuC = cpuC->subColMatrix(0, endCol);
-    MatrixPtr subGpuB = gpuB->subColMatrix(0, endCol);
-    MatrixPtr subGpuC = gpuC->subColMatrix(0, endCol);
-    cpuA->sumOfSquares(*subCpuB, *subCpuC);
-    gpuA->sumOfSquares(*subGpuB, *subGpuC);
-  }
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, 1);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckErr(*cpuA, *outputCheck);
-}
-
-void testMatrixBinaryClassificationError(int height, int width) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuC = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
-  MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
-  MatrixPtr gpuC = std::make_shared<GpuMatrix>(height, width);
-
-  MatrixPtr cpuA2 = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuB2 = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuC2 = std::make_shared<CpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  cpuB->randomizeUniform();
-  cpuC->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  gpuB->copyFrom(*cpuB);
-  gpuC->copyFrom(*cpuC);
-  cpuA2->copyFrom(*cpuA);
-  cpuB2->copyFrom(*cpuB);
-  cpuC2->copyFrom(*cpuC);
-
-  real scale = 0.5;
-  int columnOffset = rand() % width;  // NOLINT
-
-  cpuA->binaryClassificationError(columnOffset, *cpuB, *cpuC, scale);
-  gpuA->binaryClassificationError(columnOffset, *gpuB, *gpuC, scale);
-  cpuA2->binaryClassificationError2(columnOffset, *cpuB2, *cpuC2, scale);
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckErr(*cpuA, *outputCheck);
-  MatrixCheckErr(*cpuA, *cpuA2);
-}
-
-TEST(Matrix, aggregate) {
-  for (auto height : {1, 11, 16, 32, 64, 73, 128, 200, 1024, 2345}) {
-    for (auto width : {1, 9, 16, 32, 64, 100, 512, 1000, 1024, 2453}) {
-      VLOG(3) << " height=" << height << " width=" << width;
-      testMatrixCollectBias(height, width);
-      testMatrixTernaryRowDotMul(height, width);
-
-      testMatrixSumOfSquares(height, width);
-      testMatrixBinaryClassificationError(height, width);
-    }
-  }
-}
-
-TEST(Matrix, aggregate2) {
-  for (auto height : {16, 32, 128, 512, 1024}) {
-    for (auto width :
-         {16, 32, 64, 128, 256, 512, 768, 1024, 2048, 3072, 4096}) {
-      VLOG(3) << " height=" << height << " width=" << width;
-
-      int endCol = rand() % width;  // NOLINT
-      testMatrixSumOfSquares(height, width, endCol);
-    }
-  }
-}
-
 void testMatrixAddAtOffset(int height, int width1, int width2) {
   MatrixPtr cpuInput = std::make_shared<CpuMatrix>(height, width1);
   MatrixPtr cpuOutput = std::make_shared<CpuMatrix>(height, width2);